Prompt:
AI psychosis has reached epidemic proportions yet those disciplines most likely to provide some remedies have been curiously quiet on the topic. Except Active Inference. The active inference community is actively researching how world model building by individuals and groups results in both promising and dead end belief systems. Which researchers in the Active Inference space are most concerned with AI psychosis as a topic? What does the field have to say on the topic?
Excerpt from Perplexity’s response:
Active Inference frames psychosis as failed free energy minimization, where agents build flawed generative models leading to rigid or "dead-end" beliefs. For AI psychosis, it implies chatbots' sycophantic reinforcement creates loops amplifying low-precision priors, akin to human delusional systems.
The field emphasizes remedies like precision-weighted evidence integration to avoid entrapment in maladaptive models, with potential for AI design (e.g., VERSES AI's federated inference) to mitigate risks. Research calls for empirical tests of AI exposure on belief dynamics.
